Tempotron Learning

Definition

Tempotron learning is an online gradient-based supervised learning rule for spiking neuron models that implement a binary classification of multi-neuronal spike patterns. A neuronal classifier whose synaptic efficacies are controlled by the tempotron learning rule is referred to as tempotron. The tempotron implements a binary decision rule: A spike pattern is classified as target pattern if the tempotron fires at least one output spike in response to the pattern. If the neuron remains silent, the pattern is classified as nullpattern. The tempotron is trained on a training set consisting of labeled target and null patterns. When the tempotron misclassifies an input pattern during training, i.e., when its output does not match the desired response specified by the label, the tempotron learning rule adjusts each of the neuron’s synaptic efficacies according to its contribution to the maximal postsynaptic membrane potential: increasing it when the desired response is to fire...